FusionBERT: Multi-View Image--3D Retrieval via Cross-Attention Visual Fusion and Normal-Aware 3D Encoder

Name: FusionBERT: Multi-View Image--3D Retrieval via Cross-Attention Visual Fusion and Normal-Aware 3D Encoder
Uploaded: 2026-10-01T13:28:12.823741+00:00
Duration: 286 s
Description: The episode discusses FusionBERT, a novel framework for multi-view image-to-3D retrieval. The hosts explain how it uses cross-attention for visual fusion and a normal-aware 3D encoder to create unified representations of images and 3D models. The key takeaway is that combining these methods leads to stronger, more accurate matching capabilities than current state-of-the-art models.

The episode discusses FusionBERT, a novel framework for multi-view image-to-3D retrieval. The hosts explain how it uses cross-attention for visual fusion and a normal-aware 3D encoder to create unified representations of images and 3D models. The key takeaway is that combining these methods leads to stronger, more accurate matching capabilities than current state-of-the-art models.

Open the full episode